Piano World Home Page Forums Our Most Popular Forums Digital Pianos - Electronic Pianos - Synths & Keyboards New stage pianos from Yamaha!

If you can record directly that would be great. (Perhaps you can record to a USB jump drive / stick / key). Some of the newer Yamahas have this feature. I find that recording using a (cheap) mic or video camera mic are less than useful, when it comes to evaluating the sound quality!

That actually IS large for ROM. Don't forget these will be compressed samples with real time decompression going on. The sample libraries on your computer are not compressed.

We need to compare apples to apples, in terms of size here. RAM / uncompressed v. ROM / compressed.

Also, don't forget that some of the modeling magic will result in the samples being smaller, I suspect.

Except the forms of compression they've used in the past to make these smaller are stretching, looping, and fewer velocity layers. Sorry, I stopped wanting to hear that tired kind of compression years ago. Maybe this DP line will finally be different, but who knows?


No offense, but you have an awful lot of faith in a bunch of guys who never met a microscopic ROM they didn't love.

I'm always surprised when I see the insides of a DP - I expect to find ferrite core RAM and diode ROM.

Oh, if it were only that simple. The thing is, I really want a state-of-the-art DP, something you just turn on and play, not some software running on a PC (though that has it's place). Now and then I'm asked to recommend DPs for various people and installations, and most people just aren't technically up to a SW piano setup. Most can barely turn on the powerstrip to the PA.

V-Piano was looking good for a while, though it's definitely not very portable, and then I realized that modeling takes multiple generations to get anywhere near right. At least Pianoteq gives free updates, but who knows what Roland will offer? In the non-modeled category no one offers a large enough sample.


Well, that's the point, nothing I want is being offered by anyone. Don't you want a DP that is good enough to record with? Don't a lot of people here want that? This isn't just pointless bitching, I know manufacturers watch sites like this to at least some degree. And we all know they can build it, they just won't for some reason. I've asked KAWAI James about it pointedly and repeatedly, but he's been totally mum on the subject.

Are there any other industry types here that can weigh-in on this subject?

dewster -

Exactly! That's what I'm looking for as well, although of course I'm willing to use the computer when I need to record. However, really, I'd like to simply press the power button and play. Not deal with multiple program loads, and configuring files, and click, click, click with the mouse.

What is your opinion of the Yamaha CLP-380?

Lawrence

Haven't heard it, but 5 velocity layers is sort of barely getting there. 8 key up + 8 key down = 16 is what I would consider to be minimum for a sampled piano (as would many I suspect if they were buying a PC sample set 5 years ago). The real question is the size of the sample, which Yamaha probably doesn't want to tell you. They idiotically consider that kind of info to be proprietary - funny, those who sell PC sample sets know it is a huge (some would say THE) selling point and advertise it all over the place.

OK, I'll say it again: I can buy a 4GB USB flash drive at newegg for $10. Why are there no 4GB AP samples to be found in any DPs???? It simply can't be economics at this point...

Lossless compression (e.g. FLAC, TrueHD) achieves about a compression ratio of 3:1 with 16bit material. So even if the ROMs are compressed by using lossless compression, those 142mb would just become 450mb. That's still really low compared to what PC software does. Ivory comes with 40GB worth of sample data. Of course it's possible that they use lossy compression for the ROMs, but that comes with its own problems. Lossy compression works well enough for a final song. But if you actually compress the original PCM data, which must still be processed, mixed together etc, the loss by lossy compression hurts much more. Furthermore I would guess that typical DP CPUs aren't fast enough to uncompress 128 (polyphony) lossy samples at the same time, when using really high quality codecs like MP3 or AAC. So if ROMs are compressed, it's probably a rather simple compression algorithm. Which means that either quality suffers, or the compression ratio can't be too great.


As was said before, you need a certain speed which may or may not be delivered by standard off the shelf flash components. Furthermore, not only the storage itself must be more capable, also the data bus and the main processing chip must be able to handle the higher load. It's possible that there are simply no standard hardware chips available yet which have the power/functionality to handle such large samples without getting too hot. We don't want fans in our DPs, do we? But I don't really know.

I do think, though, that top-of-the-line DPs are expensive enough that manufacturers should really pull out all the stops. If temperature is a problem, they could implement large external heatsinks and connect them to the CPU with heatpipes. Going the extra mile would also give manufacturers a valid reason for the much higher price of top-of-the-line DPs compared to budget DPs.

If I can record myself playing my S6 sample on my S90XS and then record myself playing a real S6 and guarantee you all that none of you could pick by ear which is which.... Whats the issue ?
Big sample , small sample , no sample ....if it works don't be picky. Because then you've stopped becoming a musician and started being a technician.

Yeah, you need quite a bit of watts to keep a decent processor and memory running even these days so heat, noise and reliability is really a problem. Think about fitting a 500 watt power supply unit into your DP with a quad core 2,5 GHz processor, 4-8 GB of DDR2/DDR3 ram, and some kind of storage device for 40-80 GB. In addition to that you need the motherboard, high quality audio chip + the normal gear of a DP.

So you've got 500 watts of power that needs lots of cooling but has to be very silent, and the whole system has to be robust in any kind of use for more than 5 years (physical shocks, humidity and temperature changes...) - no wonder why the DP manufacturers go under the fence here.

PC sample sets are not compressed. It doesn't make any sense to do that when memory and hard drive space cost so little.


What do you base that assertion on? What did you mean exactly by "hurts much more"?

Depending on the how the sample data is encoded in the compressed domain, it might be entirely possible to do simple transformations like scaling single samples, and interpolating and mixing multiple samples before bulk decompression (i.e. if compression / decompression is largely a linear operation). But that's educated guessing on my part, and I don't think compression and its associated computational overhead are necessary at all the minute cheap flash is available (~4 years ago).


Flash bandwidth is not an issue.

Interpolating between two samples for detuning purposes, and also interpolating between two velocity layers for layer blending purposes, would require 4 Flash accesses per sample. For 128 polyphony at 44.1kHz:

4 * 128 * 44100 = 22,579,200 samples per second, or one sample every 44.29 ns. Two banks of relatively slow Flash operating at 70 ns could easily do this.


As you can see from the calculation above, the processor and memory have to handle 22.6 million samples per second peak. If the processor runs at 1 GHz, this gives 44 clocks per sample, which should be plenty (if the operating system overhead can be managed, which an RTOS is designed to do).

Also, you can easily buy a DP or synth that will do 128 polyphony. And I would submit that looping requires the same or more computation than simple sample playback, so we're already there in terms of processor power - we've been there for a while now.


Amen brother, amen.


Take a look at this product for ballpark processing power and memory / storage bandwidth:

WD TV

It streams and decompresses HD video. Sells for $80. No heat issues.


Yes, this is what I really don't get. I believe they could charge something of a premium for a recording quality DP, and I believe there is a demand for one, yet for some reason it doesn't exist.

You could do it with 5 watts, 10 if you're sloppy. No need for fans at those levels of power dissipation. No need for tons of RAM or multiple cores either.

I'm not really talking PC stuff here, but a 1 GHz ARM or equivalent, which is much more power efficient than even the Atom.

Ever heard the soundtrack to "To Live and Die in LA"? On the track "The Red Stare" single piano notes are played very slowly with the pedal down. If you listen through headphones you can hear the sympathetic resonance of the entire piano. You can hear the felt hitting the strings when he releases the pedal to damp the strings. I bet I could tell the difference between your real piano and your S90SX if you were playing that.

That's the kind of realism I demand in a DP, I want it to fool me playing anything.


Well, that's the point, it doesn't work for me. I really wish it did.


We all become technicians to one degree or another when we are confronted with the purchase of a DP. I bet the size of the piano sample in your shiny new S90SX was a selling point for you - of so, then who's being the technician now?

Now you've embarrassed me because I didn't remember that track which considering I played with Jack and Nick for nearly a year many moons ago got me to go find it. You might have picked the one track which would show up any DP in contemporary music but I could play it on a Roland Fantom G8 without you knowing as its got a separate sample for the kind of resonance effect your hearing on that track. Seriously I didn't know the size of the sample before I got the S90XS I just played it and it sounded good to me.

Wow! That is seriously cool. Please do elaborate.


Anyone else that is interested can listen to the beginning of it here:

The Red Stare


The G8 resonance may be nice (I haven't heard it) but I would probably be able to hear the looping, which would kill it for me. They list the wave memory as "256MB (16-bit linear equivalent)" which I would guess unfairly counts stretching and looping. It's not clear how much of that memory is dedicated to the AP voices.


I would submit that it sounds good because it is much larger than normal for a DP (though still quite small by PC sample standards). An even bigger sample would sound even better.

Not much to elaborate on my years of being a DX7 /Prophet /Fairlight geek are kind of a blur these days. I played with quite a few acts both touring and in the studio after my band broke up in 1984. Then I settled down married had kids and ended up in studio and stage production and the dreaded advertising business. I've been working on movie and tv music as well as stage shows and doing the odd album production since. Just got the band back together for a reunion tour in September and that's reminded me why I never missed it ;-)

But where do you store those 40 GB samples like in Synthogy Ivory or Best Service Galaxy and read them fast enough to produce real-time no-latency playback?

Never said they were. I was talking about DP ROMs.


If you take lossy sources and process/mix/modify them, things can only get worse. There are many examples in the audio and video processing world that show this. E.g. if you post process lossily compressed video, things can get ugly very fast. Lossy compression is great, but I'd use it only for the final step. At least that's my opinion. I don't have a scientific proof.


One sample consists of 4 bytes, though (16bit * stereo). Also there is *always* some overhead. Plus read spead must be noticeably faster than the theoretical needs, because you want a safety net.


I don't have enough information about the internal design of a typical DP. If you loop samples a lot, you have to read much less data. Yes, the amount of data going to the DAC/outputs is the same. But the CPU does not only process, it's also responsible for controlling the data transfer, I think. So higher data transfers from ROM to CPU will also stress the CPU more. It's also possible that the CPUs typically used for DPs are simply not built in such a way that they can read data that fast from any source. I don't know that, though...


I'm well aware of these devices. The Sigma Designs 8625 chip typically used by these HD video players gets quite hot. There are heat issues with some such players out there. And many of them have fans inside. I have one of these devices at home - fanless. I can barely touch it when it's playing back HD video. Also I believe the 8625 would be much too slow for a DP situation. Ok, it does decode HD video and that needs a lot of performance, but it's all done hard wired and not by the general processing part of the chip. The general processing part of the 8625 is so slow that it can't move more than about 10MB/s over a Gigabit Ethernet port, IIRC.

On a SSD :

PQI 32 GB Solid State HD

Much, much higher read data speeds (lower access time, higher sustained bandwidth), much more durable, and lower power consumption than spinning platter-based drives. Cost is ~$100 (and dropping like a rock).

Sequential Access - Read: Up to 170MB/s

You do get the low power consumption with that though and if combined with a low power consuming processor - it could work without noisy fans in there.

I think you would still need atleast some kind of fast ram for this system to work? That 170 MB/s is not enough, but it could be enough to feed the memory buffer of about 1GB or something. Why not just put 8GB-16GB of memory and load all needed samples to memory on start-up?

But, if you would want any kind of adjustable effects in order to mimic the acoustics of a concert hall for example, it would require a lot more computing power or some kind of DSP.

DP and synth manufacturers often count a single stereo note playing as polyphony 2. Even in that case 128 should be fine for almost any kind of playing on a DP.

The sample banks would be 16 or more bits wide, and my example was with old flash just to show that it was possible long ago. These days the sample playing would most likely be placed, in whole or in chunks, in dynamic RAM, from a SSD. It's more efficient to use DMA for this transfer, which is dedicated hardware that doesn't depend on the processor core.


Looping is a two stage process, and the transition from pure sample playback to pure loop playback is generally a crossfade. Which means that during the crossfade both processes are running concurrently. Clearly even this double load is not too much for the current inexpensive processors used in DPs.